CN116108742A - Low-voltage transformer area ultra-short-term load prediction method and system based on improved GRU-NP model - Google Patents
Low-voltage transformer area ultra-short-term load prediction method and system based on improved GRU-NP model Download PDFInfo
- Publication number
- CN116108742A CN116108742A CN202211595583.8A CN202211595583A CN116108742A CN 116108742 A CN116108742 A CN 116108742A CN 202211595583 A CN202211595583 A CN 202211595583A CN 116108742 A CN116108742 A CN 116108742A
- Authority
- CN
- China
- Prior art keywords
- model
- gru
- data
- load
- improved
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000004140 cleaning Methods 0.000 claims abstract description 10
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims abstract description 6
- 238000010219 correlation analysis Methods 0.000 claims abstract description 5
- 238000002474 experimental method Methods 0.000 claims abstract description 5
- 238000012545 processing Methods 0.000 claims description 24
- 238000005457 optimization Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 18
- 238000010606 normalization Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 230000006872 improvement Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 8
- 230000000306 recurrent effect Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 230000001276 controlling effect Effects 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Geometry (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Primary Health Care (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Computer Hardware Design (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The method and the system for predicting the ultra-short-term load of the low-voltage transformer area based on the improved GRU-NP model firstly carry out data cleaning on original load data, introduce a distance correlation coefficient to carry out characteristic correlation analysis and reject the characteristic with lower correlation; then normalizing the data set and dividing the data set into a training set and a testing set according to the proportion of 8:2; building GRU optimized based on RMSProp algorithm and NP model optimized by using Optuna; then, searching the optimal weight fused by the GRU model and the NP model by using a multi-layer perceptron MLP, and obtaining a mixed model according to the optimal weight; and loading the data to be predicted into a trained GRU-NP model to predict the load value at the future moment. Aiming at ultra-short-term load prediction of a low-voltage transformer area, an improved GRU-NP model is adopted to predict a load value of 15 minutes in the future, and experiments show that the prediction precision of multi-characteristic load data of the low-voltage transformer area is effectively improved.
Description
Technical Field
The invention belongs to the field of load prediction in an electric power environment, and particularly relates to a low-voltage transformer area ultra-short-term load prediction method and system based on an improved GRU-NP (gated recurrent unit-nerve propset) model.
Background
The gating recurrent neural network (gated recurrent neural network) is proposed to better capture the dependency of the time step distance in the time series, and controls the flow of information through a gate that can be learned. Among them, the gate-controlled loop unit (GRU, gated recurrent unit) is a common gate-controlled loop neural network. It introduces the concept of reset gate and update gate, thus modifying the way the hidden state is calculated in the recurrent neural network.
The Neural propset is a decomposable time sequence prediction model developed by a core data science team of Meta company, is a user-friendly time sequence prediction tool realized based on PyTorch, and has the main functions of the 2018 open source prediction tool propset, and is mainly used for time sequence data analysis. The Neural propset consists of different components such as trends, seasonal, auto-regressions, other regressions, etc., with three major model components being trend, seasonal, and holiday.
The multi-layer perceptron (MLP, multilayer Perceptron) is also called an artificial neural network (ANN, artificial Neural Network), which may have multiple hidden layers in between, in addition to input and output layers, and the MLP model may be implemented using different algorithms. For the hybrid model, the MLP model is implemented using back propagation, as this approach can give the predicted results higher accuracy.
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides a low-voltage transformer area ultra-short-term load prediction method and system based on an improved GRU-NP model, and the accuracy of low-voltage transformer area multi-feature load data prediction is improved.
The invention adopts the following technical scheme.
The ultra-short-term load prediction method for the low-voltage transformer area based on the improved GRU-NP model comprises the following steps:
step 1, carrying out normalization processing on an original data set, calculating a distance correlation coefficient of temperature, humidity and rainfall characteristics and load data, selecting the strong correlation characteristics as external factors, and cleaning and completing the load data;
step 2, building a GRU model based on RMSProp optimization, training the model by using the load data obtained in the step 1 and the strongly related characteristic data, building an Optuna super-parameter-based optimizing NP model, and training the NP model by using the load data obtained in the step 1 and combining the time sequence characteristics;
step 3, searching the optimal weight of the combination of the GRU model and the NP model by using a three-layer MLP network, thereby obtaining an improved GRU-NP model;
and 4, predicting the prediction results of the GRU model and the NP model obtained in the step 2 by using an improved GRU-NP model to obtain a future load value, and taking the future load value as a final prediction result.
Preferably, step 1 specifically includes:
step 1.1, carrying out Min-Max normalization on an original data set, wherein the Min-Max normalization formula is as follows:
in the method, in the process of the invention,
x represents the data before the processing and,
x' represents the data after the processing and,
X max represents the maximum value of the data and,
X min representing a minimum value of the data;
step 1.2, introducing a distance correlation coefficient to perform characteristic correlation analysis on temperature, humidity and rainfall in the original data set, and calculating by using a distance correlation coefficient formula;
step 1.3, calculating a temperature and load data distance correlation coefficient, a humidity and load data distance correlation coefficient, a rainfall and load data distance correlation coefficient according to the step 1.2, wherein the distance correlation coefficient is 0-0.2, which means that the two sequences are irrelevant or weakly relevant, and selecting a characteristic with the distance correlation coefficient larger than 0.2 as an external factor to participate in training of the GRU model;
and 1.4, cleaning the load data, filling in the blank value and removing the abnormal value.
Preferably, in step 1.2, the distance correlation coefficient formula is as follows:
in the method, in the process of the invention,
mu represents the sequence of load data,
v represents a characteristic data sequence of temperature, humidity and rainfall,
dcov (μ, v) represents the covariance of μ and v,
dcorr (μ, v) represents the distance correlation coefficient of μ and v.
Preferably, step 2 specifically includes:
step 2.1, dividing the load data obtained in the step 1 and the strong correlation characteristic data into a training set and a testing set according to a proportion;
step 2.2, building a GRU model based on RMSProp optimization, adding a Dropout layer to reduce the risk of overfitting, wherein each gating cycle unit comprises: an update gate, a reset gate; the updating gate is used for controlling the output of the hidden state at the previous moment and the degree to which the input of the hidden state at the current moment is brought into the hidden state at the current moment; the reset gate is used for controlling the hidden state output of the previous moment to flow into the candidate hidden state of the current moment;
step 2.3, setting GRU model super parameters, and optimizing the GRU model by using a RMSProp algorithm;
and 2.4, inputting the training set into the GRU model based on RMSProp optimization for training, and obtaining a trained GRU model.
Preferably, the parameter update procedure using RMSProp algorithm is as follows:
updating the state component s t Calculating a random gradient grad t A prime weighted moving average of the square terms of (a) is given by:
s t ＝γs t-1 +(1-γ)grad t ⊙grad t
in the method, in the process of the invention,
t represents a step of time in time,
gamma represents the attenuation coefficient, is a super parameter, takes the value (0, 1),
grad t representing a random gradient of the gradient,
s t-1 representing the state component of the current time instant,
S t representing a previous time state component;
updating argument x t The learning rate of each element in the independent variable is readjusted by element operation, and the formula is as follows:
in the method, in the process of the invention,
epsilon represents a constant added to maintain the stability of the numerical value, and the value is 10 -6 ，
x t-1 The input data representing the previous moment in time,
x t input data representing the current time of day is displayed,
l r representing the global learning rate.
Preferably, step 2 further comprises:
step 2.5, performing super-parameter optimization on the NP model by using Optuna; selecting a change point list, the number of change points, the regular trend and the learning rate as a super-parameter list to be optimized, and selecting a hysteresis order through a comparison experiment, wherein the hysteresis order is finally fixed to be 4;
and 2.6, building an NP model by adopting an Optuna optimized super-parameter list, wherein the formula is as follows:
y(t)＝T(t)+S(t)+H(t)+E(t)
in the method, in the process of the invention,
t (T) represents a trend term, S (T) represents a season term, H (T) represents holidays, special event effects, E (T) represents a random error term;
and 2.7, inputting the training set into the NP model for training to obtain a trained NP model.
Preferably, step 3 specifically includes:
step 3.1, constructing a three-layer MLP network, wherein the formula is as follows:
f(x)＝G(b 2 +w 2 *f(w 1 *X+b 1 )))
in the method, in the process of the invention,
x represents the input load data and,
f (x) represents the predicted value of the output,
w 1 the weight coefficients representing the hidden layers of the first layer,
w 2 the weight coefficients representing the hidden layer of the second layer,
b 1 the bias term representing the hidden layer of the first layer,
b 2 representing the bias term of the hidden layer of the second layer,
f(w 1 *x+b 1 ) Representing the result of the first layer hidden layer calculation,
g represents an activation function, which is a ReLU activation function;
and 3.2, training the GRU model and NP model prediction results obtained in the step 2 by using an MLP network to obtain an optimal weight coefficient, and combining the models by using the optimal weight coefficient to obtain an improved GRU-NP model.
A low-voltage area ultra-short-term load prediction system based on an improved GRU-NP model comprises: the system comprises a data processing module, a model building module, a model improvement module and a prediction module, wherein:
the data processing module is used for carrying out normalization processing on the original data set, calculating the distance correlation coefficient of temperature, humidity and rainfall characteristics and the load data, selecting the strong correlation characteristics as external factors, and cleaning and completing the load data;
the model building module is used for building a GRU model based on RMSProp optimization, training the model by using the load data obtained by the data processing module and the strongly related characteristic data, building an NP model based on Optuna super-parameter optimization, and training the NP model by combining the load data with time sequence characteristics;
the model improvement module is used for searching the optimal weight of the combination of the GRU model and the NP model by using the three-layer MLP network so as to obtain an improved GRU-NP model;
the prediction module is used for predicting the prediction results of the GRU model and the NP model by using the improved GRU-NP model to obtain a load value of one minute in the future, and the load value is used as a final prediction result.
A terminal comprising a processor and a storage medium; wherein:
the storage medium is used for storing instructions;
the processor is operative in accordance with the instructions to perform steps in accordance with a low voltage region ultra-short term load prediction method based on an improved GRU-NP model.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of a method for low voltage region ultra-short term load prediction based on an improved GRU-NP model.
The invention has the advantages that compared with the prior art,
according to the improved GRU-NP model, the weak correlation characteristic is removed by calculating the distance correlation coefficient through characteristic engineering, RMSProp (root mean square prop) and Optuna are introduced to optimize the GRU model and the NP model respectively, and the optimal weight for fusion is found through a multi-layer perceptron, so that the temperature characteristic advantage of the GRU model and the time sequence, season and holiday influence components of the NP model are combined, and the ultra-short-term load prediction precision of a low-voltage station area is effectively improved.
Compared with the traditional neural network (GRU, LSTM, BP), the hybrid model has the timing sequence characteristics which are not possessed by the hybrid model, compared with the Prophet model (the previous generation of the NP model), the method has the advantages that the prediction precision is improved by introducing the autoregressive network, and compared with the single NP model, the method has the temperature characteristics. In addition, the NP model super-parameters are optimized by introducing Optuna, so that the model prediction precision is higher.
Drawings
FIG. 1 is a flow chart of a low-voltage area ultra-short-term load prediction method based on an improved GRU-NP model;
FIG. 2 is a graph of temperature versus load data;
FIG. 3 is a graph showing the comparison of the predicted result and the true value of the improved GRU-NP model;
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. The embodiments described herein are merely some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art without making any inventive effort, are within the scope of the present invention.
The data set used in the method is a real load data set in a 2016-year electrotechnical modeling competition A, the Time range is from 1 month and 1 day in 2012 to 12 months and 31 days in 2013, load data sampled every 15 minutes at intervals, the total load value is about 3.5 ten thousand times, the data is preprocessed firstly, then characteristic engineering is carried out to remove weak correlation characteristics, and a training set and a test value are constructed by taking Time step as 4. Then respectively constructing a 4-feature GRU model and an NP model, adopting MSE, RMSE, MAE, R 2 As an evaluation index, a model training result shows that the 4-feature GRU model and the NP model do not perform well, so that the ultra-short-term load prediction method of the low-voltage transformer area based on the improved GRU-NP model is provided, a distance correlation coefficient is calculated for carrying out correlation analysis, RMSProp, optuna is introduced for respectively optimizing the GRU model and the NP model, and finally the advantages of the GRU model and the NP model are combined in a mode of distributing optimal weights through a multi-layer perceptron.
Example 1.
The method for predicting the ultra-short-term load of the low-voltage transformer area based on the improved GRU-NP model is shown in fig. 1, and comprises the following specific operation steps:
step 1, carrying out normalization processing on an original data set, calculating a distance correlation coefficient of temperature, humidity and rainfall characteristics and load data, selecting the strong correlation characteristics as external factors, and then cleaning and completing the load data.
Step 1.1, carrying out Min-Max normalization on an original data set, eliminating dimension influence, enabling a model training gradient to drop faster, and adopting the following Min-Max normalization formula:
wherein X is data before processing, X' is data after processing, X max Is the maximum value of the data, X min Is the minimum value of the data;
step 1.2, introducing a distance correlation coefficient to replace a pearson correlation coefficient to perform characteristic correlation analysis on temperature, humidity and rainfall in an original data set, and overcoming the defect that the pearson correlation coefficient is not necessarily independent between two variables even if 0, wherein the distance correlation coefficient is expressed as follows:
in the method, in the process of the invention,
mu represents the sequence of load data,
v represents a characteristic data sequence of temperature, humidity and rainfall,
dcov (μ, v) represents the covariance of μ and v,
dcorr (μ, v) represents the distance correlation coefficient of μ and v.
Step 1.3, as shown in Table 1, calculating a temperature and load data distance correlation coefficient dcorr t 0.6951, the relationship between temperature and load data is shown in FIG. 2, and the relationship between humidity and load data distance is a coefficient dcorr h 0.1827, a rainfall-to-load data distance correlation coefficient dcorr r 0.1159, the correlation coefficient is 0-0.2, and the two sequences can be regarded as uncorrelated or weakly correlated, so that the characteristic temperature with the distance correlation coefficient larger than 0.2 is selected as an external factor to participate in the training of the GRU model;
table 1 distance correlation coefficient of three features and load data
And 1.4, cleaning the load data, filling in the blank value and removing the abnormal value.
And 2, building a GRU model based on RMSProp optimization, training the model by using the mode of combining the load data obtained in the step 1 with temperature factors, building an Optuna super-parameter optimizing NP model, and training the model by using the load data obtained in the step 1 and time sequence characteristics. And updating the internal weight parameters of the NP model to obtain the trained NP model.
Wherein the timing characteristics include: year, month, quarter, holiday, etc.
Step 2.1, dividing a training set and a testing set according to 8:2, and constructing training Batch data according to the Time step of 4 and the Batch Size of 128;
step 2.2, building a GRU model based on RMSProp optimization, adding a Dropout layer to reduce the risk of overfitting, wherein each gating cycle unit comprises: an update gate, a reset gate; updating door z t For controlling the hidden state output h of the previous time t-1 t-1 Hidden state input x at current time t t A hidden state h brought to the current instant t t Reset gate r t For controlling the hidden state output h of the previous time t-1 t-1 How much flows into the candidate hidden state at the current time t
In (a) and (b);
step 2.3, setting GRU model super parameters, and optimizing the GRU model by using a RMSProp algorithm; the RMSProp algorithm is a self-adaptive learning rate method, adopts an exponential weighted moving average, solves the problem that the learning rate is too small to approach an optimal solution in the later period of iteration, and introduces a parameter updating process of the RMSProp algorithm as follows:
updating the state component s t I.e. calculating random gradient grad t A prime weighted moving average of the square terms of (a), the formula is described as follows:
s t ＝γs t-1 +(1-γ)grad t ⊙grad t
in the method, in the process of the invention,
t represents a step of time in time,
gamma represents the attenuation coefficient, is a super parameter, takes the value (0, 1),
grad t representing a random gradient of the gradient,
s t-1 representing the state component of the current time instant,
s t representing the state component at the previous time.
Updating argument x t In the independent variablesThe learning rate of each element is readjusted by element operation, and the formula is described as follows:
in the method, in the process of the invention,
epsilon represents a constant added to maintain the stability of the numerical value, and the value is 10 -6 ，
x t-1 The input data representing the previous moment in time,
x t input data representing the current time of day is displayed,
lr denotes the global learning rate.
Step 2.4, inputting a training set into the GRU MODEL based on RMSProp optimization for training to obtain a trained MODEL MODEL_GRU, and updating weight parameters w in the MODEL;
step 2.5, performing super-parameter list optimization on the NP model by using Optuna; selecting a change point list (changepoints), the number (n_changepoints), trend regularization (trend_reg) and learning rate (learning_rate) as super-parameter lists to be optimized, and selecting a hysteresis order (n_lags) through a comparison experiment, wherein the hysteresis order (n_lags) is finally fixed to be 4; the comparison experiment refers to selecting different hysteresis orders (n_lags), and comparing the evaluation indexes of R2 values, mae, mse and map.
Step 2.6, building an NP model by adopting an Optuna optimized super-parameter list; the NP model consists of a number of modules, the formula being described as:
y(t)＝T(t)+S(t)+H(t)+E(t)
in the method, in the process of the invention,
y (t) represents the output of the NP model,
t (T) represents a trend term, S (T) represents a season term, H (T) represents holidays, special event effects, E (T) represents a random error term;
step 2.7, inputting the training set into an NP MODEL for training, and updating weights to obtain a trained MODEL MODEL_NP;
step 3, searching the optimal weight of the GRU model combined with the NP model by using a three-layer MLP network, obtaining a mixed model according to the optimal weight,
step 3.1, constructing a multi-layer perceptron (MLP) network, which is formed by a plurality of fully connected layers, wherein the hybrid model uses three layers of MLP, and the formula is described as follows:
f(x)＝G(b 2 +w 2 *f(w 1 *x+b 1 )))
in the method, in the process of the invention,
x represents the input load data and,
f (x) represents the predicted value of the output,
w 1 weight coefficient representing first layer hidden layer
w 2 The weight coefficients representing the hidden layer of the second layer,
b 1 the bias term representing the hidden layer of the first layer,
b 2 representing the bias term of the hidden layer of the second layer,
f(w 1 *x+b 1 ) Representing the result of the first layer hidden layer calculation,
g represents the activation function, which is a ReLU activation function,
f represents a first layer activation function, and a nonlinear factor is introduced into the nerve cells, so that the nerve network can approach the nonlinear function, and a ReLU activation function is selected;
and 3.2, training the GRU model and NP model prediction results obtained in the step 2 by using an MLP network to obtain an optimal weight coefficient, and combining the model with the improved GRU-NP model by using the optimal weight coefficient for load prediction.
And 4, inputting the prediction results of the GRU model and the NP model obtained in the step 2 into an improved GRU-NP model for prediction to obtain a load value of one minute in the future, and taking the load value as a final prediction result. The prediction results are shown in fig. 3.
The present embodiment preferably divides the training set and the test set by 8:2 for the prediction results of the GRU model and the NP model.
Quantitative secondary Mean Square Error (MSE), root Mean Square Error (RMSE), mean Absolute Error (MAE), R 2 Score four evaluation criteria.
Example 2.
A low-voltage area ultra-short-term load prediction system based on an improved GRU-NP model comprises: the system comprises a data processing module, a model building module, a model improvement module and a prediction module, wherein:
the data processing module is used for carrying out normalization processing on the original data set, calculating the distance correlation coefficient of temperature, humidity and rainfall characteristics and the load data, selecting the strong correlation characteristics as external factors, and cleaning and completing the load data;
the model building module is used for building a GRU model based on RMSProp optimization, training the model by using the load data obtained by the data processing module and the strongly related characteristic data, building an NP model based on Optuna super-parameter optimization, and training the NP model by combining the load data with time sequence characteristics;
the model improvement module is used for searching the optimal weight of the combination of the GRU model and the NP model by using the three-layer MLP network so as to obtain an improved GRU-NP model;
the prediction module is used for predicting the prediction results of the GRU model and the NP model by using the improved GRU-NP model to obtain a load value of one minute in the future, and the load value is used as a final prediction result.
Example 3.
Embodiment 3 of the present invention provides a computer-readable storage medium.
A computer readable storage medium having stored thereon a program which when executed by a processor performs the steps in low voltage region ultra-short term load prediction based on an improved GRU-NP model according to one embodiment of the present invention.
The detailed steps are the same as those of the low-voltage area ultra-short-term load prediction method based on the improved GRU-NP model provided in embodiment 1, and will not be described in detail here.
Example 4.
The embodiment 4 of the invention provides electronic equipment.
An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in the method for low voltage region ultra-short term load prediction based on the improved GRU-NP model according to embodiment 1 of the present invention when the program is executed.
The detailed steps are the same as those of the low-voltage area ultra-short-term load prediction method based on the improved GRU-NP model provided in embodiment 1, and will not be described in detail here.
The invention has the advantages that compared with the prior art,
according to the improved GRU-NP model, the weak correlation characteristic is removed by calculating the distance correlation coefficient through characteristic engineering, RMSProp (root mean square prop) and Optuna are introduced to optimize the GRU model and the NP model respectively, and the optimal weight for fusion is found through a multi-layer perceptron, so that the temperature characteristic advantage of the GRU model and the time sequence, season and holiday influence components of the NP model are combined, and the ultra-short-term load prediction precision of a low-voltage station area is effectively improved.
Compared with the traditional neural network (GRU, LSTM, BP), the hybrid model has the timing sequence characteristics which are not possessed by the hybrid model, compared with the Prophet model (the previous generation of the NP model), the method has the advantages that the prediction precision is improved by introducing the autoregressive network, and compared with the single NP model, the method has the temperature characteristics. In addition, the NP model super-parameters are optimized by introducing Optuna, so that the model prediction precision is higher.
The present disclosure may be a system, method, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for causing a processor to implement aspects of the present disclosure.
The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.
The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.
Computer program instructions for performing the operations of the present disclosure can be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, c++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present disclosure are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information of computer readable program instructions, which can execute the computer readable program instructions.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (10)
1. The low-voltage station ultra-short-term load prediction method based on the improved GRU-NP model is characterized by comprising the following steps of:
step 1, carrying out normalization processing on an original data set, calculating a distance correlation coefficient of temperature, humidity and rainfall characteristics and load data, selecting the strong correlation characteristics as external factors, and cleaning and completing the load data;
step 2, building a GRU model based on RMSProp optimization, training the model by using the load data obtained in the step 1 and the strongly related characteristic data, building an Optuna super-parameter-based optimizing NP model, and training the NP model by using the load data obtained in the step 1 and combining the time sequence characteristics;
step 3, searching the optimal weight of the combination of the GRU model and the NP model by using a three-layer MLP network, thereby obtaining an improved GRU-NP model;
and 4, predicting the prediction results of the GRU model and the NP model obtained in the step 2 by using an improved GRU-NP model to obtain a future load value, and taking the future load value as a final prediction result.
2. The method for predicting ultra-short term load of low voltage transformer area based on improved GRU-NP model of claim 1,
the step 1 specifically comprises the following steps:
step 1.1, carrying out Min-Max normalization on an original data set, wherein the Min-Max normalization formula is as follows:
in the method, in the process of the invention,
x represents the data before the processing and,
x' represents the data after the processing and,
X max represents the maximum value of the data and,
X min representing a minimum value of the data;
step 1.2, introducing a distance correlation coefficient to perform characteristic correlation analysis on temperature, humidity and rainfall in the original data set, and calculating by using a distance correlation coefficient formula;
step 1.3, calculating a temperature and load data distance correlation coefficient, a humidity and load data distance correlation coefficient, a rainfall and load data distance correlation coefficient according to the step 1.2, wherein the distance correlation coefficient is 0-0.2, which means that the two sequences are irrelevant or weakly relevant, and selecting a characteristic with the distance correlation coefficient larger than 0.2 as an external factor to participate in training of the GRU model;
and 1.4, cleaning the load data, filling in the blank value and removing the abnormal value.
3. The method for predicting ultra-short term load of low voltage transformer area based on improved GRU-NP model of claim 2,
in step 1.2, the formula of the distance correlation coefficient is as follows:
in the method, in the process of the invention,
mu represents the sequence of load data,
v represents a characteristic data sequence of temperature, humidity and rainfall,
dcov (μ, v) represents the covariance of μ and v,
dcorr (μ, v) represents the distance correlation coefficient of μ and v.
4. The method for predicting ultra-short term load of low-voltage transformer area based on improved GRU-NP model of claim 3,
step 2, specifically comprising:
step 2.1, dividing the load data obtained in the step 1 and the strong correlation characteristic data into a training set and a testing set according to a proportion;
step 2.2, building a GRU model based on RMSProp optimization, adding a Dropout layer to reduce the risk of overfitting, wherein each gating cycle unit comprises: an update gate, a reset gate; the updating gate is used for controlling the output of the hidden state at the previous moment and the degree to which the input of the hidden state at the current moment is brought into the hidden state at the current moment; the reset gate is used for controlling the hidden state output of the previous moment to flow into the candidate hidden state of the current moment;
step 2.3, setting GRU model super parameters, and optimizing the GRU model by using a RMSProp algorithm;
and 2.4, inputting the training set into the GRU model based on RMSProp optimization for training, and obtaining a trained GRU model.
5. The method for predicting ultra-short term load of low voltage transformer area based on improved GRU-NP model of claim 4,
the parameter update procedure using RMSProp algorithm is as follows:
updating the state component s t Calculating a random gradient grad t A prime weighted moving average of the square terms of (a) is given by:
s t ＝γs t-1 +(1-γ)grad t ⊙grad t
in the method, in the process of the invention,
t represents a step of time in time,
gamma represents the attenuation coefficient, is a super parameter, takes the value (0, 1),
grad t representing a random gradient of the gradient,
s t-1 representing the state component of the current time instant,
S t representing a previous time state component;
updating argument x t The learning rate of each element in the independent variable is readjusted by element operation, and the formula is as follows:
in the method, in the process of the invention,
epsilon represents a constant added to maintain the stability of the numerical value, and the value is 10 -6 ，
x t-1 The input data representing the previous moment in time,
x t input data representing the current time of day is displayed,
lr denotes the global learning rate.
6. The method for predicting ultra-short term load of low voltage transformer area based on improved GRU-NP model of claim 5,
step 2 further comprises:
step 2.5, performing super-parameter optimization on the NP model by using Optuna; selecting a change point list, the number of change points, the regular trend and the learning rate as a super-parameter list to be optimized, and selecting a hysteresis order through a comparison experiment, wherein the hysteresis order is finally fixed to be 4;
and 2.6, building an NP model by adopting an Optuna optimized super-parameter list, wherein the formula is as follows:
y(t)＝T(t)+S(t)+H(t)+E(t)
in the method, in the process of the invention,
t (T) represents a trend term, S (T) represents a season term, H (T) represents holidays, special event effects, E (T) represents a random error term;
and 2.7, inputting the training set into the NP model for training to obtain a trained NP model.
7. The method for ultra-short term load prediction of a low voltage region based on an improved GRU-NP model of claim 6,
the step 3 specifically comprises the following steps:
step 3.1, constructing a three-layer MLP network, wherein the formula is as follows:
f(x)＝G(b 2 +w 2 *f(w 1 *X+b 1 )))
in the method, in the process of the invention,
x represents the input load data and,
f (x) represents the predicted value of the output,
w 1 the weight coefficients representing the hidden layers of the first layer,
w 2 the weight coefficients representing the hidden layer of the second layer,
b 1 the bias term representing the hidden layer of the first layer,
b 2 representing the bias term of the hidden layer of the second layer,
f(w 1 *x+b 1 ) Representing the result of the first layer hidden layer calculation,
g represents an activation function, which is a ReLU activation function;
and 3.2, training the GRU model and NP model prediction results obtained in the step 2 by using an MLP network to obtain an optimal weight coefficient, and combining the models by using the optimal weight coefficient to obtain an improved GRU-NP model.
8. A low voltage domain ultrashort term load prediction system based on an improved GRU-NP model using the method of any one of claims 1-7, comprising: the model comprises a data processing module, a model building module, a model improvement module and a prediction module, and is characterized in that:
the data processing module is used for carrying out normalization processing on the original data set, calculating the distance correlation coefficient of temperature, humidity and rainfall characteristics and the load data, selecting the strong correlation characteristics as external factors, and cleaning and completing the load data;
the model building module is used for building a GRU model based on RMSProp optimization, training the model by using the load data obtained by the data processing module and the strongly related characteristic data, building an NP model based on Optuna super-parameter optimization, and training the NP model by combining the load data with time sequence characteristics;
the model improvement module is used for searching the optimal weight of the combination of the GRU model and the NP model by using the three-layer MLP network so as to obtain an improved GRU-NP model;
the prediction module is used for predicting the prediction results of the GRU model and the NP model by using the improved GRU-NP model to obtain a load value of one minute in the future, and the load value is used as a final prediction result.
9. A terminal comprising a processor and a storage medium; the method is characterized in that:
the storage medium is used for storing instructions;
the processor is operative according to the instructions to perform the steps of the low voltage region ultra-short term load prediction method based on the modified GRU-NP model according to any one of claims 1-7.
10. A computer readable storage medium having stored thereon a computer program, which when executed by a processor, implements the steps of the improved GRU-NP model based low voltage district ultra short term load prediction method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211595583.8A CN116108742A (en) | 2022-12-13 | 2022-12-13 | Low-voltage transformer area ultra-short-term load prediction method and system based on improved GRU-NP model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211595583.8A CN116108742A (en) | 2022-12-13 | 2022-12-13 | Low-voltage transformer area ultra-short-term load prediction method and system based on improved GRU-NP model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116108742A true CN116108742A (en) | 2023-05-12 |
Family
ID=86258841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211595583.8A Pending CN116108742A (en) | 2022-12-13 | 2022-12-13 | Low-voltage transformer area ultra-short-term load prediction method and system based on improved GRU-NP model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116108742A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116628473A (en) * | 2023-05-17 | 2023-08-22 | 国网上海市电力公司 | Power equipment state trend prediction method based on multi-factor neural network algorithm |
| CN116667326A (en) * | 2023-05-30 | 2023-08-29 | 淮阴工学院 | Electric automobile charging load prediction method |
-
2022
- 2022-12-13 CN CN202211595583.8A patent/CN116108742A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116628473A (en) * | 2023-05-17 | 2023-08-22 | 国网上海市电力公司 | Power equipment state trend prediction method based on multi-factor neural network algorithm |
| CN116667326A (en) * | 2023-05-30 | 2023-08-29 | 淮阴工学院 | Electric automobile charging load prediction method |
| CN116667326B (en) * | 2023-05-30 | 2024-02-23 | 淮阴工学院 | Electric automobile charging load prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sangiorgio et al. | Robustness of LSTM neural networks for multi-step forecasting of chaotic time series | |
| CN111970163B (en) | Network flow prediction method of LSTM model based on attention mechanism | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN116108742A (en) | Low-voltage transformer area ultra-short-term load prediction method and system based on improved GRU-NP model | |
| TW201947510A (en) | Insurance service risk prediction processing method, device and processing equipment | |
| CN110910004A (en) | Reservoir dispatching rule extraction method and system with multiple uncertainties | |
| CN109063939A (en) | A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network | |
| Ayvaz et al. | An integrated LSTM neural networks approach to sustainable balanced scorecard-based early warning system | |
| WO2022246843A1 (en) | Software project risk assessment method and apparatus, computer device, and storage medium | |
| Laaroussi et al. | Deep learning framework for forecasting tourism demand | |
| Girimurugan et al. | Application of deep learning to the prediction of solar irradiance through missing data | |
| AU2021106200A4 (en) | Wind power probability prediction method based on quantile regression | |
| CN117633184A (en) | Model construction and intelligent reply method, device and medium | |
| Lei et al. | A novel time-delay neural grey model and its applications | |
| CN116303786B (en) | Block chain financial big data management system based on multidimensional data fusion algorithm | |
| CN117394350A (en) | Short-term power load prediction method, device, equipment and medium | |
| CN116957136A (en) | Pavement performance prediction method and device based on time sequence deep learning | |
| Osorio et al. | Distributed evolutionary algorithms with adaptive migration period | |
| CN115392441A (en) | Method, apparatus, device and medium for on-chip adaptation of quantized neural network model | |
| Lian et al. | A fuel sales forecast method based on variational Bayesian structural time series | |
| Loseva et al. | Ensembles of neural networks with application of multi-objective self-configurable genetic programming | |
| Gaber et al. | Hourly electricity price prediction applying deep learning for electricity market management | |
| CN111931994A (en) | Short-term load and photovoltaic power prediction method, system, equipment and medium thereof | |
| CN112183814A (en) | Short-term wind speed prediction method | |
| Basterrech et al. | Solar irradiance estimation using the echo state network and the flexible neural tree |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |